ResearchBenchmarksTranscript-level Impact Assessment on RealCBTFollow1.341Mean ImpactGPT-OSS1.172521.216261.261.30374Jul 3, 2026Evaluation ResultsMethodMethodLinksMean ImpactImpact Std DevMean Baseline Perturbation (ε)GPT-OSSFramework=MCoTFramework=MCoT2026.071.3410.50.053MistralFramework=MCoTFramework=MCoT2026.071.2560.4420.04Gemma3Framework=MCoTFramework=MCoT2026.071.1790.690.061